ES2969140T3 - Device for optical inspection of parisons - Google Patents

Abstract

A device (1) for the optical inspection of parisons (2) comprises: an illuminator (3) configured to emit a beam of light directed at a parison (2) located in an inspection position (10); a detector (4) configured to capture an image (10) of the parison (2) interposed between the illuminator (3) and the detector (4), where the illuminator (3) includes an emission polarizing filter (32) configured to generate a beam of polarized light, and where the detector (4) includes a receiving polarizing filter (41) configured to receive the beam of polarized light. (Automatic translation with Google Translate, without legal value)

Description

DESCRIPTION

Device for optical inspection of parisons

Technical field

This invention relates to a device for the optical inspection of parisons. Lines for the production of plastic containers, specifically bottles, commonly comprise a molding machine, configured to form parisons of<p>E<t>(polyethylene terephthalate) and a blow molding machine configured to blow the parisons into molds to make the containers. Some parisons may present defects, for example, due to irregular thicknesses, holes, blisters or foreign bodies; These defects must be detected through quality control aimed at quickly eliminating defective parisons from the production line.

Previous technique

Known from patent document EP2976204B1 is an optical parison control system configured to check parisons while they are transported by a conveyor towards a collection container; The system comprises a camera and a light source, arranged such that the light source illuminates the parisons from behind and the camera captures an image of each parison from the front.

Other inspection systems are disclosed in the following patent documents: US2017/129157A1, DE102006047150A1, US2018/311883A1. A limitation of prior art optical inspection systems is that they limit quality control to image features that are visible to the camera (such as blistering, for example) but cannot detect internal material defects. as irregular residual stresses in the PET polymer chains.

Additionally, prior art systems detect defective parisons based on similarity to images of other defective parisons stored in a database. Therefore, these systems must be initialized with a database containing all possible defects that must be detected; However, it is rare to find such a complete and exhaustive database because defects vary widely and defective parisons constitute only a small percentage of manufactured parisons.

Therefore, generally speaking, prior art systems have limited reliability in identifying defective parisons.

Disclosure of the invention

The objective of this disclosure is to provide a device and method for optical inspection of parisons to overcome the disadvantages of the prior art mentioned above. Another objective of this disclosure is to provide a method for processing an image of a parison to overcome the disadvantages of the prior art mentioned above.

These objectives are fully achieved, according to this disclosure, by the device for optical inspection of parisons, by the method for optical inspection of parisons and by the method for processing an image of a parison as characterized in the attached claims.

More specifically, this disclosure relates to a device for optical inspection of parisons (or preforms) or other objects made of plastic material (such as caps or capsules, for example) or metal (such as caps); In this regard, it is understood that the rest of this description, made with reference to parisons, is also applicable to other objects made of plastic or metallic material. The term "parison" is used to designate an intermediate product in the production process of plastic containers such as beverage bottles. More specifically, parisons are formed by molding the plastic material (typically injection or compression molding) and, at a later stage, expanded by blow molding to produce the finished containers. The parisons are made of plastic material, preferably PET (polyethylene terephthalate).

The device for optical inspection of parisons according to this disclosure comprises an illuminator. The illuminator includes a light source configured to emit a beam of light directed at a parison located in an inspection position.

In one embodiment, the device includes an inspection pocket configured to receive the parison in the inspection position. In another embodiment, the inspection bag configured to receive the parison in the inspection position is part of a line that, among other things, includes the device. The device comprises a detector. The detector includes a camera. The camera (i.e., detector) is configured to capture an image of the parison located at the inspection position.

In one embodiment, the light source is configured to emit the light beam continuously. In one embodiment, the light source is strobe and is configured to emit the light beam at predetermined emission intervals (each emission interval corresponds to the time required to feed a parison to the inspection position).

In one embodiment, the camera is configured to capture the image at predetermined capture intervals (each capture interval corresponds to the time required to feed a parison to the inspection position); if the light source is strobe, the emission intervals correspond (i.e., are equal to) the capture intervals.

In the inspection position, the parison is operatively interposed between the illuminator and the detector. In this way, the illuminator illuminates the parison from a first side (for example, from the front), while the detector captures an image of the parison from a second side, opposite to the first (for example, from behind); The detector thus captures a back-illuminated image of the parison.

Preferably, the illuminator (or device) includes an emission polarizing filter (or first polarizing filter). The emission polarizing filter is configured to intercept the light beam emitted by the light source. The emission polarizing filter is configured to generate a corresponding polarized light beam from the light beam emitted by the light source.

Preferably, the detector (or inspection device) includes a receiving polarizing filter (or second polarizing filter).

Preferably, in the inspection position, the parison is operatively interposed between the emission polarizing filter and the receiving polarizing filter.

The receiving polarizing filter is configured to receive the polarized light beam. More specifically, the receiving polarizing filter is configured to receive the beam of light polarized by the emission polarizing filter and refracted by the parison and to generate a second beam of polarized light. The camera thus receives the second beam of polarized light.

Since the material from which the parison is made (preferably PET) is characterized by birefringence, when the polymer chains inside it are oriented and subjected to tension, the polarized light (from the first polarizing filter) that falls on it is refracts according to refractive indices that differ according to the tension. More specifically, the light rays that hit the parison are decomposed into two rays that oscillate in perpendicular planes and the second polarizer allows only part of the components to pass, placing them in the same plane and creating interferences; Therefore, areas subjected to the same voltage will have the same interferences and, therefore, the same color, while areas subjected to different voltages will have different colors. In this way, the camera captures an image showing a color pattern that represents an internal distribution of stress in the parison.

It is provided that the first and second polarizing filters are arranged in parallel planes. The first polarizing filter (or emission filter) is intended to be a linear polarizing filter. In particular, the first polarizing filter (or emission filter) is configured to polarize light in a first polarization direction. The axis of the parison, in the inspection position, is expected to be oriented parallel to the first polarization direction (of the emission filter). Here the "parison axis" is understood to be the central symmetry axis of the parison, around which the lateral wall of the parison extends.

The polarizing filter (or receiving filter) is intended to be a linear polarizing filter; The second polarizing filter (or receiving filter) is configured to polarize light in a second polarization direction. Preferably, both the first and second polarizing filters are linear polarizing filters. In one embodiment, the first polarization direction and the second polarization direction are parallel to each other (and in this case, the resulting light is white). In one embodiment, the first polarization direction and the second polarization direction are different from each other. In particular, the first and second polarization directions may define an angle between 5° and 50°; for example, the first and second polarization directions may be perpendicular to each other (and in this case, the resulting light is black). In another example, the first and second polarization directions may define an angle of 45°. In another example, the first and second polarization directions may define an angle of 30°. In another example, the first and second polarization directions may define an angle of 15°.

In other embodiments, the first and/or second polarizing filters are circular polarizing filters.

In other embodiments, the first and second polarizing filters are not provided and therefore the camera captures a backlit monochrome image of the parison (with light and dark areas).

It is planned that the parisons will be fed individually, that is, one after the other, at the inspection post. For example, it is envisaged that the inspection device is part of an apparatus (which may itself be the subject of the present disclosure), which includes a conveyor configured to transport in succession the parisons to be inspected along an inspection path. (in which, within the inspection tour, the inspection position is defined). The conveyor may include a suction belt that defines a plurality of openings and is configured to contact an upper edge of the parisons, to create a negative pressure (i.e., a vacuum) in the internal cavity of the parisons. The suction belt is then configured to support the parisons through said negative pressure. The suction belt is configured to move the parisons along the inspection path and place them, one after another, in the inspection position. The illuminator and the emission polarizing filter may be arranged on a first side of the inspection path (i.e., of the suction belt); the camera and the receiving polarizing filter may be arranged on a second side of the inspection path (i.e., of the suction belt), opposite to the first side; Therefore, the parison, held by the suction belt, is interposed between the emitting polarizing filter and the receiving polarizing filter.

In one embodiment, the device comprises a processing unit. The processing unit is connected to the detector.

The processing unit includes memory that contains reference data. The processing unit includes a processor, programmed to process the image captured by the detector based on the reference data, in order to derive a diagnostic indication regarding the parison defect.

In one embodiment, the processor may include one or more criteria for identifying bad parisons, based on one or more threshold values stored in memory. For example, the processor can be configured to identify a defective parison in the case where the brightness of the image captured by the camera is above a predetermined threshold value (in effect, an image with high brightness indicates a parison with very thin walls. ) and/or in the event that the image presents a sharp color transition, above a certain threshold).

In a preferred embodiment, the processing unit is configured to process the image captured by the detector (based on the reference data), in order to derive values of a plurality of image characteristics from the captured image; This process encodes the image code (transforming or compressing the image according to a predetermined algorithm, or one derived by the processing system in a self-learning stage). Therefore, the processing unit is configured to assign a value to each feature of the plurality of image features. In one embodiment, encoding the image includes reducing the size of the image (for example, the plurality of image features may include 500 features).

These image features represent the image. Therefore, the processing unit is configured to extract from the image a plurality of values assigned to a corresponding plurality of image features and to reduce the image to these representative (or significant) values of the image.

In one embodiment, the processing unit is configured to process the values of the plurality of image features (based on the reference data) to derive a diagnostic indication regarding the parison defect. In one embodiment, the processing unit is configured to classify the image based on the values of the plurality of image features; For example, for each image feature, the memory may include one or more values typical of a good (i.e., defect-free) parison and one or more values typical of a defective parison, and the processing unit may be configured to identify A parison is defective if at least one value of an image feature is reasonably close (above a certain threshold) to the respective typical value of the defective parison, and/or if a certain combination (e.g., a multiplication) of features of the image is reasonably close (above a certain threshold) to a reference value for that combination, typical of a defective parison.

For example, image characteristics may include the presence or absence of certain colors or color combinations, and/or the presence or absence of a certain symmetry and/or light intensity at certain points; The processing unit can be configured to identify a parison as defective if the image has a certain color or combination of colors, or has (or does not have) a certain symmetry, or the light intensity at certain points is greater or less than a value. threshold.

In one embodiment, the processing unit is configured to generate a reconstructed image from the values of the plurality of image features (and based on the reference data). In one embodiment, the processing unit is configured to derive the diagnostic indication regarding the parison defect based on comparing the image captured by the detector with the image that is reconstructed (based on the reference data).

More specifically, the processing unit is configured to compare the image captured by the detector with the reconstructed image and to derive a similarity parameter that represents a similarity between the image captured by the detector and the reconstructed image. The processing unit is configured to compare the similarity parameter with a predetermined threshold value (which may be part of the reference data) and to derive the diagnostic indication based on comparing the predetermined threshold value with the similarity parameter. For example, the processing unit may be configured to identify a parison as good if the similarity parameter is above a certain similarity threshold. Therefore, the processing unit is configured to identify a parison as good if the reconstructed image is sufficiently similar to the initial image captured by the camera.

In effect, the processing unit is trained to process the image (i.e., encode it, derive the values of the plurality of image features and generate the reconstructed image therefrom) on the basis of parisons that are good, that is, free of defects. The processing unit may be trained using a self-learning system, as described below, or may be a commercially available pretrained unit. If the parison is good, the processing unit is able to process your image correctly and generate a reconstructed image that is similar to the original image; conversely, if the parison is defective, the processing unit, which is trained on good parisons, cannot reconstruct the image correctly and, as a result, generates a reconstructed image that is significantly different from the original.

Therefore, it is expected that the processing unit will be trained with samples of objects of a single typology (for example, defect-free typology); then, the processing unit will be able to distinguish between at least two typologies of objects (for example, typology with defects and typology free of defects).

In one embodiment, the device comprises a self-learning system.

The self-learning system is, in one embodiment, integrated into the processing unit. The self-learning system is, in one embodiment, connected to the processing unit. The self-learning system is, in one embodiment, connected to memory.

The self-learning system is configured to receive as input a plurality of images captured by the detector for a corresponding plurality of parisons. The self-learning system is configured to encode each image of the plurality of images captured by the detector for a corresponding plurality of parisons (based on the reference data) to derive from each image of the plurality of images corresponding values of a plurality of image characteristics. Preferably, the self-learning system is configured to encode the images based on a predetermined criterion (which may itself be part of the reference data).

The self-learning system is configured to generate for each image of the plurality of images a corresponding image reconstructed from the corresponding values of the plurality of image features (and based on the reference data).

The self-learning system is configured to compare each image of the plurality of images with the corresponding reconstructed image in order to derive, for each image, a similarity parameter that represents a similarity between the image captured by the detector and the corresponding reconstructed image. .

The self-learning system is configured to update the reference data based on the similarity parameter such that the similarity parameter is less than a predetermined threshold value (if the similarity parameter is directly proportional to the similarity between the images; in Otherwise, if the similarity parameter is inversely proportional to the difference between the images, so that it is greater than a predetermined threshold value). In one embodiment, the predetermined threshold value is itself part of the reference data. Preferably, the self-learning system is configured to update the reference data (iteratively) based on the similarity parameter for each image of the plurality of images.

In a preferred embodiment, the self-learning system is configured to update the plurality of image features (preferably in combination with the reference data) based on the similarity parameter and a predetermined threshold value. More specifically, the reference data and/or the plurality of image features are updated such that, for each image of the plurality of images, the similarity parameter is greater than a predetermined threshold value (if the similarity parameter is directly proportional to the similarity between the images).

On the other hand, if the similarity parameter is directly proportional to a difference between the images (i.e., inversely proportional to their similarity), the self-learning system is configured to update the reference data and/or image features of such that the similarity parameter is less than a predetermined threshold value.

In a preferred embodiment, the predetermined criterion includes (or is defined by) a maximum number (or a predetermined number) of features for the plurality of image features. The default criterion ensures that the system does not simply perform an identification function on the captured image, regardless of the parison defect.

Thus, in a preferred embodiment, the self-learning system is configured to determine both image characteristics and reference data based on images captured by the detector.

This system works particularly well if the captured images represent good parisons; that way, the system learns to encode and reconstruct the good parisons images (i.e., the reference data to be used for encoding, the image features to be extracted (because they represent good parisons), and the reference data that will be used for reconstruction. Then, when the processing system has to encode and reconstruct a defective parison, it will fail and will generate a reconstructed image that differs significantly from the initial one.

In another embodiment, the predetermined criterion includes (or is defined by) the plurality of image features (i.e., the features for which values are extracted).

In one embodiment, the self-learning system includes classifiers. The classifiers are configured to classify the captured images based on parameters such as, for example, color, color gradient, color standard deviation in a neighborhood of one pixel, mean color in a neighborhood of one pixel or others. Specifically, the classifiers can be programmed to construct a histogram representative of a distribution of colors in the image and to classify the image based on a symmetry and/or uniformity of the colors in said histogram. In one embodiment, it is envisioned that the parameters may be preset in the classifiers (i.e., explicitly defined). For example, classifiers are expected to be of the "one class" type. The classifiers can be configured to select, from a plurality of preset parameters, one or more parameters to use during the inspection (this selection can be made by the classifiers during the learning stage).

In one embodiment, the classifiers are decision-making branches. In particular, "one-class" classifiers are expected to be "random forest" decision-making branches. It is observed that these classifiers can be trained (only) with images of objects without defects (since the classifier belongs to the "one class" typology).

In a preferred embodiment, the self-learning system includes neural networks. In particular, the self-learning system may include convolutional neural networks.

By "convolutional neural networks" we mean neural networks configured to encode the image through a series of convolution stages alternating with a series of pooling stages to derive the values of the plurality of image features. In the convolution stages, convolution filters (whose values are part of the reference data) are applied to the image (i.e., to the array of pixels representing each color in the image) to derive a transformed image; In the pooling stages, a dimension of the transformed image is reduced, for example, by a maximum, minimum or average mathematical operation between adjacent pixels. Therefore, the convolution and pooling stages are used to obtain the values of the plurality of image features.

In one embodiment, the neural networks are pretrained to extract (i.e., derive) values from the plurality of image features; In one embodiment, the neural networks may include known neural networks (or parts thereof) configured to identify subjects in an image (e.g., based on an "ImageNet" data set).

In one embodiment, the convolutional neural networks are configured to classify the image based on the values of the plurality of features in the image, compared to the reference values (which are part of the reference data). In one embodiment, the image is classified as representing a good parison or a bad parison; In another embodiment, the image is classified as representing a good parison or a parison having a specific defect. In one embodiment, the neural networks include "anomaly detection" classifiers or "fully connected" networks to classify the image based on the values of the plurality of image features. For example, to extract (i.e. derive image feature values), a part of a well-known data set-based network such as "ImageNet" is used, the network that includes convolution and pooling (hence a previously trained network), and to classify the image a "fully connected" network is used, trained by the self-learning system (preferably with examples of good parisons and examples of bad parisons) and which is capable of distinguishing a good parison from a bad parison. based on image feature values.

In one embodiment, the learning system includes neural networks of the type known as "Generative Adversarial Networks" (GAN); Generative Adversarial Networks include a generator and a discriminator; the generator uses the values of the plurality of image features (obtained with the convolution stage from a real image captured by the detector) to generate a corresponding reconstructed image and passes it to the discriminator; the discriminator tries to distinguish whether the image it receives is real - that is, captured by the detector - or reconstructed, and sends feedback to the generator; Based on feedback from the discriminator, the generator learns to generate reconstructed images that are as similar as possible to the real images (so that the discriminator considers them real). During training, the GAN preferably receives images of good parisons; thus, when the processing system uses the reference data (and image features) determined by the self-learning system to inspect the parisons, the discriminator recognizes as reconstructed images only those related to defective parisons.

In a preferred embodiment, the self-learning system includes an "autoencoder" convolutional neural network. In one embodiment, the reference data includes a first set of reference data and a second set of reference data. The first reference data set relates to a series of convolution stages (for example, includes a plurality of convolution filters); The second set of reference data relates to a series of upsampling stages. In particular, the second reference data set may be related to a series of deconvolution steps (for example, includes a plurality of deconvolution filters). The autoencoder neural networks are configured to extract (i.e., derive) values from the plurality of reference features of the image, based on the first set of reference data related to the series of convolution steps, and to reconstruct the image. (i.e., generating the reconstructed image) using the values of the plurality of image features, based on the second set of reference data related to the series of deconvolution steps.

In the embodiment that includes an autoencoder network, the learning system is configured to receive as input a plurality of good parison images to derive corresponding values of the image features, based on the first set of data of reference; the learning system is configured to generate a plurality of reconstructed images using the image feature values and based on the second set of reference data related to the deconvolution steps; Finally, the self-learning system is configured to compare the plurality of reconstructed images with the corresponding plurality of original images and to update the reference data (specifically, the first set of reference data or the second set of reference data or both the first as the second combined reference data sets) to minimize the difference between the original images captured by the detector and the corresponding reconstructed images. More specifically, in one embodiment, the self-learning system is configured to derive, for each image, a similarity parameter (e.g., calculated with one of the following functions, or with a combination of two or more of the following functions: norm 11, I2 norm, structural similarity index "SSIM", peak signal-to-noise ratio "PSNR", Haar wavelet-based perceptual similarity index "HaarPSI") and minimize the similarity parameter (i.e., bring it below a certain threshold) if the similarity parameter is directly proportional to a difference between the images, or vice versa, maximize it (i.e., take it above a certain threshold) if the similarity parameter is directly proportional to a similarity between the images. In the embodiment that includes an autoencoder network, the learning system is configured to update, in combination with the first and second set of reference data, the filters of the network (and, consequently, the plurality of characteristics of the image), to identify the image features that best represent images of defect-free parisons and jointly identify the convolution stages to derive them and the deconvolution stages to generate the reconstructed image using those image features. Preferably, the self-learning system comprising an autoencoder network is constrained to at least one predetermined criterion (i.e., a constraint) when encoding the image; for example, the criterion may consider a maximum number of image features that can be identified for that plurality of image features. Thus, in practice, the encoding operation constitutes a compression of the image.

Thus, in one embodiment, the self-learning system is configured to learn reference data and image features that the processing system then uses to inspect parisons. In the embodiment including an autoencoder network, the self-learning system is trained with a plurality of parisons that are good (defect-free); Thus, the self-learning system determines the reference data and image features used to correctly reconstruct the images of the good parisons (i.e., with a high level of similarity between the images captured by the detector and the corresponding reconstructed images) ; When you apply those reference data and image features to a defective parison, the reconstructed image is significantly different from the corresponding captured image and the processing system, when compared, identifies the parison as defective.

It is observed that the comparison between the captured image and the reconstructed image is carried out both during learning (or training) of the network and during inspection; The similarity parameter used for training may be different from the similarity parameter that will be used during inspection. In particular, the functions "PSNR", "HaarPSI" are preferably used during inspection, but not during training. For example, it is planned to use, for training, the "SSIM" function or the I1 standard function or the I2 standard function, and for inspection the "PSNR" function or the "HaarPSI" function.

Preferably, the self-learning system (or processing system) includes a first neural (sub)network responsible for encoding the images to derive image feature values and a second neural (sub)network responsible for encoding the images to deriving the image feature values to decode the image feature values in order to reconstruct the image; the first and second neural (sub)networks are trained together. The first and second (sub)neural networks may be parts of a single neural network.

This disclosure also provides a line for manufacturing containers from thermoplastic material.

In one embodiment, the line comprises a molding machine configured to make parisons. In one embodiment, the molding machine is an injection molding machine. In one embodiment, the molding machine is a compression molding machine (which may be a rotary machine).

In one embodiment, the line comprises a thermal conditioning unit for heating and/or cooling the parisons. For ease of description, the thermal conditioning unit is hereinafter referred to as "heating oven" without limiting the scope of the disclosure. The heating oven is configured to receive the parisons exiting the molding machine and is equipped with heating means for heating the parisons.

In one embodiment, the line comprises a blow molding machine configured to receive the parisons and blow mold them into molds to manufacture the containers. Preferably, the blow molding machine is configured to receive the parisons heated in the heating oven. The oven can be integrated into the blowing machine.

In one embodiment, the blow molding machine and the parison molding machine could be located on different lines (even installed in separate factories) working together to manufacture thermoplastic material containers: in effect, the molding machine manufactures parisons that are fed to the line that includes the blowing machine. Preferably, the oven is integrated into the line that includes the blowing machine, upstream of the blowing machine to heat the parisons before blowing them.

In one embodiment, the line comprises a storage unit (which may be automatic) for receiving the parisons and storing them; The storage unit is configured to receive the parisons from the molding machine and feed them to the blow molding machine or oven.

In one embodiment, the line comprises a device for optical inspection of parisons according to one or more aspects of this disclosure. The optical inspection device is located on the line behind the molding machine. The optical inspection device is located in the line before the blowing machine. In one embodiment, the optical inspection device is located downstream of the molding machine and upstream of the oven. In one embodiment, the optical inspection device is located behind the oven and in front of the blowing machine. The fact that the device is located upstream of the blowing machine allows a defective preform to be identified before it explodes during blowing due to its abnormal stress profile. In one embodiment, the optical inspection device could be placed on the storage unit or on a conveyor that connects the storage unit to other parts of the line.

In one embodiment, the optical inspection device is located on the line such that it inspects the parisons at a temperature of between 30 and 70 degrees Celsius (preferably between 50 and 60 degrees Celsius). The parisons may be at this temperature when they come out of the molding machine, for example.

In one embodiment, the optical inspection device is located on the line such that it inspects the parisons at an ambient temperature (e.g., between 5 and 30 degrees Celsius). Parisons can be at room temperature if stored or cooled after molding. Preferably, the optical inspection device is located on the line such that it inspects the parisons at a temperature below 60 degrees Celsius (preferably below 50 degrees Celsius); in fact, at higher temperatures, parisons could be subject to deformations that could alter their stress profiles.

In one embodiment, the optical inspection device is integrated into the oven. If the optical inspection device is integrated into the oven (preferably upstream of the heaters, that is, at the entrance to the oven), it is possible to take advantage of the placement of the parisons inside the oven, singularized and ordered.

In other embodiments, the optical inspection device according to this disclosure is located offline, integrated into a high-speed screening machine or a low-speed sampling machine.

This disclosure also relates to a method for optical inspection of parisons. The optical inspection method comprises a step of emitting, with an illuminator, a beam of light directed at a parison located in an inspection position. The optical inspection method comprises a step of capturing, with a detector, an image of the parison located in the inspection position. In the inspection position, the parison is operatively interposed between the illuminator and the detector. The captured image is therefore a backlit image.

In one embodiment, the optical inspection method comprises a step of generating a polarized light beam by intercepting the light beam emitted by the illuminator in a polarizing emission filter interposed between the illuminator and the parison.

In one embodiment, the optical inspection method comprises a step of receiving the polarized light beam into a receiving polarizing filter. In the inspection position, the parison is operatively interposed between the emission polarizing filter and the receiving polarizing filter.

In one embodiment, the parison is made of a material that includes polymer chains (preferably PET). In one embodiment, the image captured by the detector includes a color pattern that represents a stress to which the PET polymer chains are subjected.

In one embodiment, the optical inspection method comprises an image processing step (using a processing system). In one embodiment, in the processing step, the image captured by the detector is processed based on reference data contained in a memory. In one embodiment, the processing step includes deriving the diagnostic indication with respect to the parison defect.

In one embodiment, the processing step includes a substep of encoding the image captured by the detector, based on reference data, to derive values of a plurality of image features from the image.

In one embodiment, the processing step includes processing the plurality of image features to derive the diagnostic indication regarding the parison defect. More specifically, in one embodiment, the processing step includes a substep of generating a reconstructed image from the values of the plurality of image features and based on the reference data. In one embodiment, the processing step includes a substep of deriving the diagnostic indication with respect to the parison defect based on comparing the image captured by the detector with the reconstructed image.

In one embodiment, the method comprises a self-learning step (performed by a self-learning system connected to the processing system). In one embodiment, the self-learning step comprises a substep of capturing a plurality of images for a corresponding plurality of parisons. In one embodiment, the self-learning step comprises a substep of encoding each image of the plurality of images based on the reference data, including deriving, from each image of the plurality of images, corresponding values of a plurality of image characteristics according to with a predetermined criterion (which may include a restriction on the maximum number of image features of the plurality of image features). In one embodiment, the self-learning step comprises a substep of generating for each image of the plurality of images, a corresponding image reconstructed from the corresponding values of the plurality of image features and based on the reference data. In one embodiment, the self-learning step comprises a substep of comparing each image of the plurality of images with the corresponding reconstructed image and deriving a corresponding similarity parameter that represents a similarity between the image captured by the detector and the corresponding reconstructed image. In one embodiment, the self-learning step comprises a sub-step of updating the reference data and/or the plurality of image features based on the similarity parameter and a predetermined threshold value. More specifically, the reference data and the plurality of image features are updated such that the similarity parameter is less than (or greater than) the predetermined threshold value.

In one embodiment, the images of the plurality of images captured by the camera during the self-learning stage represent a corresponding plurality of defect-free parisons. The fact that the self-learning system does not need to receive images of defective parisons as input is very advantageous because defective parisons are difficult to find.

Preferably, the emission polarizing filter and the receiving polarizing filter are oriented relative to each other according to a predetermined orientation (e.g., parallel or perpendicular) that is the same in the self-learning stage and in the processing stage.

In one embodiment, the method comprises a step of feeding the parisons of the plurality of parisons to the inspection position one at a time (in succession). In one embodiment, the parisons are fed according to a predetermined orientation with respect to the emission polarizing filter and with respect to the receiving polarizing filter. This orientation is the same in the self-learning stage and in the processing stage. The image of each parison of the plurality of parisons is captured when the parison is in the inspection position.

In one embodiment, the processor is configured to process randomly oriented images (e.g., rotating them to a predetermined orientation); Therefore, in one embodiment, the device is invariant with respect to the orientation of the parisons.

In one embodiment, the processing system is configured to process the parison image in real time (as soon as the image is captured). In one embodiment, when the device provides the diagnostic indication that identifies the parison as defective, the line is configured to stop and allow an operator to remove the defective parison or to itself remove the defective parison.

In another embodiment, the processing system is configured to capture the parison image for processing in post-processing mode. In this case, the system is configured to correlate each image with the respective parison in order to identify the parison whose image produced the defective parison diagnostic indication.

This disclosure also relates to a method of processing an image of a parison; It is noted that the method for processing an image of a parison, per se, is not the object of the claimed invention; However, the method for optical inspection of parisons according to the invention includes image processing, so the content of the disclosure regarding the method for processing an image of a parison is relevant to the invention. In one embodiment, the image processing method includes a step of encoding the image, which includes deriving values from a plurality of image features. In one embodiment, the image processing method includes a step of generating a reconstructed image based on the plurality of image features. In one embodiment, the image processing method includes a step of deriving the diagnostic indication regarding the parison defect based on comparing the image captured by the camera with the reconstructed image.

In one embodiment, the image processing method includes a self-learning step (in accordance with one or more aspects described in this disclosure).

This disclosure also relates to a computer program (software) comprising operating instructions configured to carry out the steps of the processing method in accordance with one or more aspects described in this disclosure (when performed by a processor, specifically the processing unit). device processing according to one or more aspects described in this disclosure).

Brief description of drawings

These and other features will become more apparent from the following detailed description of a preferred embodiment, illustrated by way of non-limiting example in the accompanying drawings, in which:

Figure 1 illustrates an optical inspection device according to this disclosure;

Figure 2 represents a process to which an image is subjected by the optical inspection device of Figure 1;

Figure 3 represents a self-learning process carried out on an image by the optical inspection device of Figure 1;

Figures 4A, 4B and 4C show, for a defect-free parison, an image captured by a camera, a reconstructed image and a comparison between the captured image and the reconstructed image, respectively; Figures 5A, 5B and 5C show, for a defective parison, an image captured by a camera, a reconstructed image and a comparison between the captured image and the reconstructed image, respectively;

Figure 6 illustrates a line for manufacturing containers of thermoplastic material, which includes the device of Figure 1.

Detailed description of preferred embodiments of the invention

With reference to the accompanying drawings, number 1 indicates an optical inspection device configured to inspect a parison 2.

The parison 2 includes a body 200 having a substantially cylindrical shape. The parison 2 (or body 200) defines an axis of symmetry A. The body 200 is therefore cylindrically symmetrical with respect to the axis of symmetry A. The parison 2 includes a closed bottom 201. The parison 2 includes a neck 202 that defines an opening. Parison 2 includes a 203 ring.

The device 1 is configured to receive a parison 2 at an inspection position 10. In one embodiment, the inspection position is defined by an inspection cavity. In one embodiment, the inspection pocket includes a support member 11 configured to hold the parison 2 (preferably supporting the ring 203).

The device 1 includes an illuminator 3. The illuminator 3 includes a light source 31. The light source 31 is configured to emit a light beam directed at a parison 2 (i.e., at an inspection position 10). The illuminator 3 includes an emission polarizing filter 32. In one embodiment, the emission polarizing filter 32 is connected to the light source 31. The emission polarizing filter 32 is configured to intercept the light beam emitted by the light source. 31 and polarize it. In this way, the parison 2 receives a beam of polarized light from the emission polarizing filter 32 and refracts it.

The device 1 includes a detector 4. The detector 4 includes a camera 41. The detector includes a receiving polarizing filter 42. In one embodiment, the receiving polarizing filter 42 is connected to the camera 41. The receiving polarizing filter 42 is configured to receive the light beam refracted by parison 2 and polarize it. Thus, the camera 41 receives the light beam polarized by the emission polarizing filter 32, refracted by the parison and further polarized by the receiving polarizing filter 42. The camera 41 is configured to capture (or acquire) an image 20 of the parison 2 .

The illuminator 3 laterally illuminates the parison 2 on a first side 200A of the body 200. The detector 4 captures a lateral image of the parison 2 on a second side 200B of the body 200, opposite the first side 200A.

Device 1 includes memory 5. Memory 5 contains reference data. More specifically, the memory 5 contains at least a first set of reference data 51 and a second set of reference data 52; In one embodiment, the first reference data set 51 and the second reference data set 52 are distinct from each other.

The device 1 includes a processor 6. The processor 6 is connected to the memory 5. The processor 6 is programmed to process the image 20 captured by the camera 41 based on the reference data sets 51, 52, in order to derive the diagnostic indication 23 with respect to the parison defect 2. More specifically, the processor 6 is programmed to perform a step 61 of encoding the image 20 based on the first set of reference data 51 to derive values of a plurality of characteristics of image 21. The processor 6 is also configured to perform a step 62 of decoding the features of the image 21, thereby generating a reconstructed image 22, based on the second set of reference data 52.

The processor 6 is then configured to perform a step 63 of comparing the reconstructed image 20 with the captured image 22 to derive a diagnostic indication 23 regarding the parison defect 2.

In one embodiment, the diagnostic indication includes an error map 25 given by a difference between the captured image 20 and the reconstructed image 22 (or vice versa). In an embodiment illustrated in the drawings, the error map 25 has uniform shading if the parison is good or patchy shading if the parison is bad.

In one embodiment, the diagnostic indication 23 includes a similarity parameter 24 whose value correlates with a degree of similarity between the captured image 20 and the reconstructed image 22. In one embodiment, the processor 6 is programmed to derive the similarity parameter 24 based on the error map 25. In one embodiment, the diagnostic indication 23 includes a binary parameter value that indicates whether the parison is good or bad (calculated, for example, by comparing the similarity parameter 24 with a predetermined threshold value ).

In one embodiment, the device 1 (or preferably the processing system) comprises a self-learning system 7. The self-learning system 7 is preferably integrated into the processor 6. The self-learning system 7 is connected to the memory 5.

The self-learning system 7 is configured to receive a plurality of captured images 20 for a corresponding plurality of parisons 2. The self-learning system 7 is preferably configured to perform the following steps for each image 20 it receives: 61 encodes the image 20 onto the based on the first reference data set 51, to derive a plurality of image features 21; 62 decode the features of the image 21, based on the second reference data set 51, to generate a reconstructed image 22; 63 comparing the reconstructed image 22 with the captured image 20 to derive a similarity parameter 24 that represents a similarity between the captured image 20 and the reconstructed image 22. 70 evaluating the similarity parameter 24 with respect to a predetermined threshold value 72 for that similarity parameter; update (iteratively) the first reference data set 51, the second reference data set 52 and the image features 21 until the similarity parameter 24 is above (or below) the threshold parameter 72.

Thus, the self-learning system 7 solves the problem of optimizing the encoding operations 61 and the decoding operations 62, where the variables are defined by the first set of reference data 51 and by the second set of reference data 52 (and , if necessary, by the set of image features 21), to minimize the similarity parameter 24, that is, bring it below a certain threshold, (or maximize it, that is, bring it above a certain threshold). Preferably, therefore, the first reference data set 51 and the second reference data set 52 are updated in combination.

Since the self-learning system 7 optimizes the encoding and decoding operations 61 and 62 with images 20 of good parisons 2, the reference data sets 51, 52 (and, if necessary, the image feature set 21) determined as a result of the optimization is such that, for good parisons 2, the difference between the captured image 20 and the reconstructed image 22 is minimal, on the other hand, since these operations are not optimized for bad parisons 2, the reconstructed image 22 for a defective parison is significantly different from the captured image 20 and the processor 6 (recognizing that difference) generates a diagnostic indication 23 meaning that the parison is defective.

Preferably, the steps 61 of encoding, 62 decoding, 63 comparing, 70 evaluating and updating the reference data sets 51, 52 (and, if necessary, the image features 21) are performed iteratively by the self-learning 7 for each image 20 in succession (i.e., all iterations necessary to minimize or maximize the similarity parameter 24 are performed first for a first parison 2, then for a second parison 2, and so on). In one embodiment, the self-learning system could also perform a first iteration in which it performs the steps 61 of encoding, 62 decoding, 63 comparing and 70 evaluating all images 20; then, from the similarity parameters 24 obtained for all parisons 2, it updates the reference data sets 51, 52 (and, if necessary, the image features 21) and continues with a second iteration in which re-performs the steps 61 of encoding, 62 decoding, 63 comparing and 70 evaluating all images 20, and so on.

This disclosure also relates to a line 100 for manufacturing containers of thermoplastic material, for example, bottles.

Line 100 comprises a molding machine 101 configured to make (i.e., mold) parisons 2. In one embodiment, the molding machine 101 is a rotary machine. The line 100 also comprises a heating oven 102 configured to receive the molded parisons 2 and heat them. Line 100 comprises a blow molding machine 103 configured to blow mold the parisons 2 to manufacture the containers. In one embodiment, the blow molding machine 103 is a rotary machine.

Preferably, the line 100 includes a first transfer carousel 106 configured to transfer the parisons 2 from the molding machine 101 to the heating oven 102. Preferably, the line 100 includes a second transfer carousel 107 configured to transfer the parisons 2 from the heating oven 102 to the blow molding machine 103. In one embodiment, the line 100 includes a storage unit 104 for storing the molded parisons 2 before they are blow molded. In one embodiment, line 100 includes a parison orientation device 105 configured to orient parisons 2 exiting and/or entering storage unit 104. In one embodiment, line 100 includes a conveyor 108 configured to transport parisons. 2 inside and/or outside the storage unit 104. The conveyor 108 feeds the parisons 2 from the storage unit 104 to the heating oven 102.

Claims (14)

1. A device (1) for the optical inspection of parisons (2), comprising: - an illuminator (3) that includes a light source (31) configured to emit a beam of light directed towards a parison (2) located in an inspection position (10); - a detector (4) that includes a camera (41) configured to capture an image (20) of the parison (2) located in the inspection position (10), in which the parison (2), in the inspection position (10), is operatively interposed between the illuminator (3) and the detector (4), wherein the illuminator (3) includes a polarizing emission filter (32) configured to intercept the light beam emitted by the light source (31) and generate a polarized light beam, and wherein the detector (4) includes a polarizing receiving filter (42) configured to receive the polarized light beam, so that the parison (2), in the inspection position (10), is operatively interposed between the emission polarizing filter (32) and the filter receiving polarizer (42); - a self-learning system; - a processing unit that includes: a memory (5) that includes reference data sets (51, 52); a processor (6) programmed to process the image (20) captured by the detector (4) based on the reference data sets (51, 52), in order to derive a diagnostic indication (23) with respect to a defect of the parison (2);, so that the diagnostic indication distinguishes between a typology of objects with defects and a typology of objects free of defects, the processing unit being trained, by the self-learning system, through samples of objects of the typology free of defects, in which the processing unit is configured to: - processing the image (20) captured by the detector (4) based on the reference data sets (51, 52), to derive from the image (20) values of a plurality of image characteristics (21); - processing the values of the plurality of image features (21) to derive the diagnostic indication (23) with respect to the parison defect (2); - generating an image (22) reconstructed from the values of the plurality of image features (21) and based on the reference data sets (51, 52); - deriving the diagnostic indication (23) relative to the parison defect (2) based on comparing the image (20) captured by the detector (4) with the reconstructed image (22). 2. The device (1) according to claim 1, wherein the self-learning system (7) is configured to: - receiving as input a plurality of images (20) captured by the detector (4) for a corresponding plurality of parisons (2); - processing each image (20) of the plurality of images (20) captured by the detector (4) based on the reference data sets (51, 52), in order to derive for each image (20) corresponding values for the plurality of image characteristics (21) based on a predetermined criterion; - generating for each image (20) of the plurality of images (20) a corresponding reconstructed image (22), based on the reference data sets (51,52), using the corresponding derived values for the plurality of image features (twenty-one); - comparing each image (20) of the plurality of images (20) captured by the detector (4) with the corresponding reconstructed image (22) to derive, for each image (20) of the plurality of images (20), a parameter corresponding similarity (24) that represents a similarity between the image (20) captured by the detector (4) and the corresponding reconstructed image (22); - for each image (20) of the plurality of images (20), updating the reference data sets (51, 52) based on the similarity parameter (24) and a predetermined threshold value (72) for the similarity parameter similarity (24) . 3. The device (1) according to claim 2, wherein the self-learning system (7) is configured to update the plurality of image features (21) based on the similarity parameter (24) and the threshold value default (72). 4. The device (1) according to claim 2 or 3, wherein the predetermined criterion includes a maximum number of image features (21) for the plurality of image features (21). 5. The device (1) according to any of claims 2 to 4, wherein the self-learning system (7) includes convolutional neural networks configured to encode images through a series of convolution stages alternating with a series of pooling steps to derive values from the plurality of image features. 6. The device (1) according to any of the preceding claims, wherein the emission polarizing filter (32) is a linear filter, configured to polarize light in a first polarization direction. 7. The device (1) according to any of the preceding claims, wherein the receiver-polarizing filter (42) is a linear filter, configured to polarize light in a second polarization direction. 8. The device (1) according to any of the preceding claims, wherein the parisons (2) are made of PET and wherein the image captured by the detector includes a color pattern that represents a voltage at which the polymer chains of PET are subjected. 9. A line (100) for manufacturing containers of thermoplastic material, comprising - a molding machine (101) configured to manufacture parisons (2); - a blow molding machine (103) configured to receive the parisons (2) and blow them into molds to manufacture the containers; - a device (1) for optical inspection of parisons (2) according to any of claims 1 to 7, in which any of the following conditions is verified: i) the optical inspection device (1) is operatively located downstream of the molding machine (101); ii) the optical inspection device (1) is operatively located upstream of the blow molding machine (103). 10. A method for optical inspection of parisons, comprising the following steps: - emit a beam of light directed towards a parison (2) located in an inspection post (10), by means of an illuminator (3) that includes a light source (31); - capture, with a detector (4) that includes a camera (41), an image (20) of the parison (2) located at the inspection post (10), in which the parison (2), at the inspection post inspection (10), is operatively interposed between the illuminator (3) and the detector (4); - generate a beam of polarized light by intercepting the beam of light emitted by the illuminator (3) on a polarizing emission filter (32) interposed between the light source (31) and the parison (2); - receiving the polarized light beam in a receiving polarizing filter (42), interposed between the parison (2) and the camera (41), in which, in the inspection position (10), the parison (2) is interposed operatively between the emission polarizing filter (32) and the receiving polarizing filter (41); - image processing (20), in which the processing step includes the following substeps: - (61) processing the image (20) captured by the detector (4) based on the reference data sets (51, 52), to derive from the image (20) values of a plurality of image characteristics (21 ); - (62) generating an image (22) reconstructed from the values of the plurality of image features (21) and based on the reference data sets (51, 52); - (63) derivation of a diagnostic indication (23) relating to a parison defect (2) based on comparing the image (20) captured by the detector (4) with the reconstructed image (22); - train the processing unit through a self-learning system, so that the diagnostic indication distinguishes between a typology of objects with defects and a typology of objects without defects, the processing unit being trained by the self-learning system, through samples of objects of the defect-free typology. 11. The method according to claim 10, wherein the self-learning step comprises the following substeps: - capturing a plurality of images (20) for a corresponding plurality of parisons (2); - processing each image (20) of the plurality of images (20) based on the reference data sets (51, 52), to derive from each image (20) of the plurality of images (20) corresponding values of a plurality of image characteristics (21) based on a predetermined criterion; - generating for each image (20) of the plurality of images (20) a corresponding reconstructed image (22) using the corresponding values of the plurality of image features (21) and based on the reference data sets (51, 52 ); - comparing each image (20) of the plurality of images (20) with the corresponding reconstructed image (22) and deriving a corresponding similarity parameter (24) that represents a similarity between the image (20) captured by the detector (4 ) and the corresponding reconstructed image (22); - updating the reference data sets (51, 52) and the plurality of image features (21) based on the similarity parameter (24) and a predetermined threshold value (72). 12. The method according to claim 11, wherein the images (20) of the plurality of images (20) captured by the camera (41) during the self-learning step are representative of a corresponding plurality of parisons (2) free of defects. 13. The method according to claim 11 or 12, comprising a step of feeding the parisons (2) of the plurality of parisons (2) to the inspection position (10) one at a time, and according to a predetermined orientation with respect to the emission polarizing filter (2) and relative to the receiving polarizing filter (42). 14. The method according to any of the preceding claims 10 to 13, wherein the emission polarizing filter (32) is a linear polarizing filter, configured to polarize light in a first polarization direction.